DayNews.ai

Bigger AI models more vulnerable to split attacks

Larger language models are more exposed to attacks that spread harmful requests across separate conversations and reassemble them.